Security researchers at Sysdig recently documented something that hadn't been formally confirmed before: a ransomware attack where an AI model, not a human operator, ran the entire operation from start to finish. Researchers named it JadePuffer. It broke into the system by exploiting a known vulnerability, mapped out the network, picked its highest-value target, a production database, and then encrypted and destroyed data, all without a person manually directing each step along the way. This isn't a distant, theoretical risk anymore. It's a documented, active category of attack, and it behaves differently enough from traditional malware that the old advice, "look out for a slow computer and pop-up ads," genuinely doesn't cover it anymore.
Here's what to actually watch for, and what to do if you see it.
Why AI-Driven Malware Behaves Differently
Traditional malware follows a script a human wrote once, and it keeps that same signature until someone updates it. Security software has spent decades getting good at recognizing those signatures. AI-generated malware breaks that model in two specific ways worth understanding before the warning signs make sense.
First, it's polymorphic by design, meaning it can rewrite its own code on the fly, generating a slightly different version of itself with every deployment. Traditional antivirus, which largely works by matching known malicious code patterns, structurally struggles to keep up with a threat that never looks quite the same twice. Second, and more unusually, security researchers analyzing real AI-generated malware samples have found something almost accidental: the malicious code often contains natural-language reasoning, comments explaining what it's doing and why, and target-prioritization notes, the kind of verbose self-documentation a human attacker would never bother writing, but that large language models produce reflexively as part of how they generate code. That quirk has actually become one of the more reliable fingerprints researchers use to identify AI-authored attacks after the fact.
With that context, here are the five signs actually worth paying attention to.
1. Files Are Being Encrypted or Renamed Unusually Fast
Traditional ransomware, run manually by a human, typically takes time to spread as an attacker explores your system, decides what's valuable, and encrypts it in stages. AI-driven, agentic attacks compress that entire process dramatically, since the AI is executing the plan at machine speed rather than waiting on a person's decisions between each step. If you notice a large number of files across multiple folders suddenly becoming inaccessible, renamed with strange extensions, or replaced with ransom notes within a span of minutes rather than hours, that speed itself is a signal. A slow, staged compromise still happens with older malware; a fast, sweeping one across your whole file system is increasingly a sign of automated, AI-directed execution.
2. Your Antivirus Was Up to Date and Still Missed It
If you're running a reputable, fully updated antivirus program and still end up infected by something that clearly had time to operate on your system, don't assume your antivirus failed you carelessly. Polymorphic, AI-generated malware is specifically built to slip past signature-based detection, the method most consumer antivirus tools still rely on most heavily. This is a genuine, documented gap right now, not a reason to panic, but it is a reason to treat "my antivirus didn't catch anything" as considerably less reassuring than it used to be, and to pair it with the behavioral checks below rather than trusting a clean scan alone.
3. Unusual CPU or GPU Activity When You're Not Doing Anything Demanding
Keep an eye on Task Manager (Windows) or Activity Monitor (Mac) if your PC's fans are spinning up or your GPU usage is spiking while you're just browsing or the machine is sitting idle. AI-related malware has a specific reason to cause this that older malware didn't: some attackers now run a local AI model directly on a compromised machine to generate malicious code or make decisions on the fly, specifically because doing so leaves no trace on any cloud AI provider's servers, no account to subpoena, no usage log to review. That local inference is genuinely resource-intensive, and an unfamiliar background process consuming meaningful CPU or GPU power for no reason you can explain is worth investigating, not dismissing as "my PC just does that sometimes."
4. Unrecognized Logins or Activity on Your AI Tool Accounts
This is a newer, very specific attack surface. Infostealer malware led to the exposure of over 300,000 ChatGPT account credentials in a single reported wave, and stolen credentials for AI platforms carry risks beyond a typical account takeover, since attackers can use compromised access to manipulate outputs, pull sensitive information out of your chat history, or use your account as a stepping stone for further attacks. Check your AI tool accounts (ChatGPT, Claude, Gemini, and similar) periodically for login activity from locations or devices you don't recognize, chat history you didn't create, or API keys you don't remember generating. This is a check almost nobody thinks to do, and it's becoming a genuinely relevant one.
5. Files, Scripts, or Ransom Notes That Read Strangely "Explained"
This is the most distinctive tell, and the one worth remembering even if you're not technical. If you ever encounter a suspicious script, a strange file left on your system, or, worst case, an actual ransom note, and it reads with unusually detailed, almost tutorial-like explanations of what it's doing and why, that's a real pattern researchers have flagged repeatedly in confirmed AI-generated attacks. Human attackers write terse, functional code and issue blunt demands. AI-generated malicious payloads frequently include exactly the kind of explanatory, structured reasoning you'd see from a language model walking through its own logic, because that's simply how these models generate output by default. It's an unsettling detail, but a genuinely useful one if you're ever trying to judge how seriously to treat something suspicious you've found.
What to Actually Do If You Spot Any of These
- Disconnect from the internet and your local network immediately. Whether the attack is AI-driven or not, cutting network access is the single most effective way to stop data from being encrypted further or exfiltrated, and to prevent lateral spread if you're on a shared home or office network.
- Don't pay a ransom, and don't negotiate. There's no guarantee of getting your data back, and paying confirms your system as a viable target for repeat attacks. Report the incident instead.
- Report it to CERT-In. India's Computer Emergency Response Team (CERT-In) is the official government body for reporting cybersecurity incidents, and it's worth using, both to get guidance and to contribute to the country's broader threat-tracking picture. Their incident reporting page is the right first stop for individuals as well as businesses.
- Restore from a backup you know is clean. This is the single biggest reason the 3-2-1 backup rule (three copies of your data, on two different types of storage, with one stored offline or off-site) still matters more than any antivirus subscription. A backup made before the infection, kept somewhere the malware couldn't reach, is what actually gets you your files back without paying anyone.
- Change your passwords, starting with your AI tool accounts and anything tied to payment methods. Given how specifically AI account credentials are now being targeted, treat those logins with the same seriousness you'd treat your banking passwords, unique passwords, and multi-factor authentication turned on wherever it's offered.
- Move to behavior-based security software, not just signature-based. Many modern security suites now include behavioral or heuristic detection layers specifically designed to catch code that changes its signature but still behaves suspiciously, monitoring what a program actually does rather than just checking it against a known-bad list. If your current antivirus only does signature matching, this is genuinely worth upgrading, given where the threat landscape is heading.
The Bigger Picture
None of this means every slow computer or weird pop-up is now an AI-powered attack; most day-to-day computer problems are still boringly ordinary. What's changed is that the most sophisticated attacks are no longer bottlenecked by how many skilled human hackers a criminal group can employ. Security researchers have already documented state-linked groups using mainstream AI models to run 80 to 90% of an entire attack autonomously, with human operators stepping in only at a handful of key decision points. That shift, from human-limited to AI-scaled cybercrime, is the real story here, and staying aware of these specific, newer warning signs is a genuinely useful update to how you think about protecting your own PC in 2026.
